NaN does not mean "I am not a number”(github.com) |
NaN does not mean "I am not a number”(github.com) |
There's probably some cutting insight into programmer culture here somewhere.
It is definitively not a number in the usual, mathematical sense -- hence the name, which comes from IEEE 754.
However, the isNumber function doesn't really check if things are numbers, it checks if they are of the Number type, which, potentially confusingly, includes things which are not numbers, including the accurately named value Not-a-Number (NaN).
That would imply then that isNumber is named incorrectly. So perhaps isNumber should read isNumericType?
Most of the time I want to test, is this a number I can use in a normal mathematical sense. As in, can I use it in subtraction, multiplication etc. I know now that I should use isFinite and isNumber, but wouldn't it be nice to have just one well named function for this?
We use NaN for masking in image processing, for instance for filtering out defective pixels. If a pixel has value NaN, it is ignored across the board.
My own interpretation? While programmers shouldn't care so much about this particular kind of semantics (instead they should understand the theory behind floats and never have to run into this issue) it might be better just to call NaNs something else, like "undefined number" or something.
Or, better, call the JavaScript type that represents IEEE-754 Binary64 format -- which includes the union of three sets (finite numbers in a given range, two infinite numbers, and two kinds of Not-a-Number values) something other than "Number".
The type should obviously have been called Float or some variant of that, and the type identification function isFloat.
1.0 / 1.0
=> Just 1.0
0.0 / 0.0
=> NothingNaN is not a number (the concept) but it is a Number (the type).
> That would imply then that isNumber is named incorrectly.
Arguably, it is Number (the type) that is named incorrectly (a type called Number that includes a value called Not-a-Number seems problematic.) Maybe "Number" should be called "MaybeNumber".
And, perhaps more importantly, the fact that JS doesn't have a convenient type membership operator for primitive types is problematic, as if there was a convenient parallel to "instanceOf" for primitive types -- or if "instanceOf" just worked for them -- then libraries wouldn't need to provide something like what underscores isNumber does at all, and then any isNumber function would probably test "is the argument a number" rather than "is the argument a Number".
The only reason this is confusing for people is because they aren't familiar with IEEE 754 floating-point numbers and their representations. But since floats are used in almost every programming language, it's something that any programmer should be familiar with. Once they learn exactly what NaN means in that context, then the confusion should disappear.
NaN is part of IEEE 754, but it is not a number.
The reason people are confused is because the JS datatype name has a small but critical disconnect from what the values it contains are.
Its confusing because NaN explicitly states not a number. Then the function explicitly states is number.
However the two are in very different contexts, yet this is not communicated when using them. It's intuitive to think of a date as a class and there for the context is implicit in isDate. I don't think this is the case for isNumber.
I think this is less about people understanding floating-point numbers more about understanding that a function starting with is implies some form of type checking.
The problem is that "Number" is the name of a JS data type, and "number" is the name of a concept, the two don't match exactly, and when you are using camel-casing for word separation, you can't distinguish the two.
The function "isNumber" checks if the thing it is applied to is a JS Number, not if it is a number.